The Annotation Ontology (AO) has proven to be a valuable resource for structuring annotations in scientific documents. We are representing elements of discourse with the AO; by using our proposed extension it is possible to mark up specific rhetorical structures and build a network of interconnected documents. The extension presented in this paper also makes it possible to represent more expressive associations across nanopublications.
Digital Libraries such as Elsevier Science Direct1 or PubMed2 store electronic
versions of scientific publications. Although these resources provide some
information retrieval mechanisms, it is still difficult to extract facts buried in the text
[
Heading towards a self-descriptive document requires a well-organized annotation structure consistent with the underlying rhetorical structure. Annotation should support not only marking segments but also making the relationships across these portions explicit. Furthermore, annotations should scaffold relations across documents. It is not enough to know the concepts in a document; it is also necessary
2 http://www.ncbi.nlm.nih.gov/pubmed/
to know how are they related [
The Annotation Ontology (AO) [
We are broadening the interoperability between SWAN and AO [
We have extended the AO with new classes and properties that facilitate making
explicit the rhetoric and discourse embedded in a scientific publication. Classes are
meant to categorize the type of structures expressed in a publication -e.g. definitions,
examples, claims, etc. From the AO, we have reused ao:Definition making it
compliant to the Meaning-of-a-tag (MOAT) [
NCIt “Watching something and taking note of what happens.” aold:Result aold:Discussion aold:Conclusion coresc:Result, “Factual statements about the outputs of an investigation” The annotation identifies a fragment corresponding to the discussion of the document coresc:Conclusion, “Statements inferred from observations & results relating to research hypothesis”
We are also proposing classes that facilitate the definition of relations between entities; this makes it possible to relate fragments within the same document or across multiple documents. Qualifiers in the AO are mapped to SKOS properties; here we interpret them as expressing a subjacent relationship between the annotated fragment/document and the topic. It is recommended to use ao:Annotation+ao:hasTopic whenever it is needed to point to examples, definitions and external links related to URIs. The ao:Qualifier/aold:OnFlyQualifier+ao:hasTopic should be used to relate the annotated fragment/document to an entity/resource, e.g. to relate “mouse” to an ontological term “ncbitaxon:10090”. Qualifiers, as proposed by AO, express only five relationships, we are proposing aold:OnFlyQualifier that extends the original ao:Annotation to model any relationship that has been defined somewhere else, typically an ontology. The property aold:definesRelation maps to the subjacent relation, e.g. owl:sameAs, between the annotated fragment/document and the topic. Relations used by a specific annotation project can be narrowed down to a set of predefined relations; they may also be open to represent any relation expressed as free text. The last scenario would probably require curation mechanisms to see whether the relation is new or can be mapped to an existing one. Fig. 1 shows two possible uses of aold:OnFlyQualifier: on the left, an annotator has identified the paper with URI “http://biotea.ws/paper1” as the same at “http://tinyurl.com/apaper”; on the right, an annotator has identified a claim and its corresponding source.
The aold:Relation extends the aold:OnFlyQualifier, it represents an annotation that brings together a pair of annotations; the ao:body of the annotation establishes the intended name for the annotation. In this way, it is possible to use aold:OnFlyQualifier for known relations while using the aold:Relation may be reserved for new ones. Two subclasses have been proposed, aold:UnidirectionalRelation for those relations where the subject and object cannot be interchanged, e.g. is_a, and aold:BidirectionalRelation for all other relations, e.g. synonyms. If aold:definesRelation is used, it defines a relation, e.g. sameAs, between the ao:body and the topic. Fig. 2 shows a graphical representation of these classes.
aold:Relation is useful in cases such as definitions in vocabularies as well as when making explicit a process described by a document. Additional information can be found at http://biotea.ws. Fig. 3 illustrates how could aold:Relation be used when expressing new ways to relate documents. An annotated fragment in a document is categorized as an “opinion” on a fragment of a second document.
We have also added two new selectors that work on RDF documents. When working with these selectors it is assumed that only one rdfs:comment will be present in the RDF element being annotated. The aold:ElementSelector extends the aos:textSelector, it identifies an exact chunk of text in the rdfs:comment while the aold:StarEndElementSelector extends the ao:StartEndSelector by identifying the start and end positions of the text being annotated in the rdfs:comment. Fig. 4 illustrates an example using the latter selector: A text from position 27 to position 38 in the “Introduction” section of the corresponding RDF representation for the paper with DOI 10.1016/SO014-5793(03)00051-6 has been annotated; the annotation body is “mus musculus”. The aold:OnFlyQualifier is here used to indicate that the annotated text corresponds to ncbitaxon:10090.
In order to support the argumentative process hidden in the text, we are reusing the
properties proposed by the SWAN [
Figures 5 and 6 present a hypothesis in a document using the definition given in a
different document; Fig. 5 uses the aold:OnFlyQualifier to establish the relation
“cites to” whereas Fig. 6 uses aold:UnidirectionalRelation. Both figures illustrate
how a fragment in the document with DOI 10.1016/SO014-5793(03)00051-6 has
been annotated as a hypothesis; this hypothesis cites a fragment in the document
PMC1435992 that corresponds to a definition that has been identified as the same
concept defined by the entity CHEBI_16113.
Nanopublications are “a set of annotations that refers to the same statement and
contains a minimum set of (community) agreed-upon annotations”; it is therefore
feasible to use the proposed extension to represent nanopublications. Consider for
instance the definitions for ontology; these vary depending on the field. One that is
commonly used comes from Gruber, “An ontology is a formal specification of a
conceptualization” [
Scientific publications annotated in the proposed way benefit from the Semantic Web and Link Open Data initiatives. Annotations and information extraction based on these publications become easier, as do sharing them and enriching them with other information also available in RDF. In this way, we facilitate the integration between literature and databases making it easier to use data available in publications for additional analysis. Our ultimate goal is to increase the pace of available scientific data by helping researchers to find information relevant to their projects. As publications are annotated, their rhetoric becomes searchable, thus researchers can better focus on those publications meeting their needs depending on what they are looking for, e.g. similar experiments based on methods, materials and protocols, or rebuttals of a particular theory. Furthermore, as annotations can be linked to databases, this enriched content can also be used for more specialized queries.
We have introduced our own description for some existing concepts in CoreSC
such as Experiment and Observation, as we wanted them to be more accurate from the
workflow laboratory perspective. We have not used the relations proposed by AZ-II
[
Our approach is compatible with the principles of nanopublications. A concept would be a minimal ao:Annotation while relations on annotations could be used to define statements that are uniquely identified by a URI. Annotations, as they are understood in nanopublications, are also possible as relations, i.e. statement, are resources that can use as the subject of an annotation. The nanopublication itself becomes concrete by using the Annotation Set proposed by AO vr. 2.0; the Annotation Set is a container of annotations that is used to organize annotations that can be referred to as a whole. Furthermore, our approach makes it possible to relate nanopublications to any other type of publication. Our approach is also compatible with other annotation models, such as MOAT and Tag Ontology9, making it easier to extend and integrate existing applications and tools. 4
Conclusions We have presented an extension to AO that facilitates modeling rhetoric and discourse in scientific publications. Our approach entails using the common practice of annotating in order to identify specifics within the text; we are also gathering relationships between the annotated and referenced objects. Categories make it easier to identify whether it is about a claim, an example, a report, etc.; some of these categories come from the AO, others are new. In addition to the terms used during the exercise, we also worked with terms such as hypothesis, conclusion and research question; these are useful when analyzing the structure of scientific publications structure. Although not analyzed here, it could also be useful for commercial
documents, since they are also related to each other as well as to external resources. How to implement our model? This is a question beyond the scope of this paper that remains open; it is one of our top priorities.
Acknowledgments. We are grateful to Paolo Ciccarese for his fruitful discussions related to Hypertag migration to AO, i.e. using annotations to express relations.